Automated Domain Modeling with Large Language Models: A Comparative Study
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Domain modeling is an essential part of software engineering and serves as a way to represent and understand the concepts and relationships in a problem domain. Typically, software engineers interpret the problem description written in natural language and manually translate it into a domain model. Domain modeling can be time-consuming and highly depends on the expertise of software engineers. Recently, Large Language Models (LLMs) have exhibited remarkable ability in language understanding, generation, and reasoning. In this paper, we conduct a comprehensive, comparative study of using LLMs for fully automated domain modeling. We assess two powerful LLMs, GPT3.5 and GPT4, employing various prompt engineering techniques on a data set containing ten diverse domain modeling examples with reference solutions created by modeling experts. Our findings reveal that while LLMs demonstrate impressive domain understanding capabilities, they are still impractical for full automation, with the top-performing LLM achieving F <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> scores of 0.76 for class generation, 0.61 for attribute generation, and 0.34 for relationship generation. Moreover, the F <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> score is characterized by higher precision and lower recall; thus, domain elements retrieved by LLMs are often reliable, but there are many missing elements. Furthermore, modeling best practices are rarely followed in auto-generated domain models. Our data set and evaluation provide a valuable baseline for future research in automated LLM-based domain modeling.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it